diff --git a/dask/array/__init__.py b/dask/array/__init__.py
index b4f5444fe8a..139dde18754 100644
--- a/dask/array/__init__.py
+++ b/dask/array/__init__.py
@@ -47,7 +47,7 @@
         from .reductions import nanprod, nancumprod, nancumsum
     with ignoring(ImportError):
         from . import ma
-    from . import random, linalg, overlap, fft
+    from . import random, linalg, overlap, fft, backends
     from .overlap import map_overlap
     from .wrap import ones, zeros, empty, full
     from .creation import ones_like, zeros_like, empty_like, full_like
diff --git a/dask/array/backends.py b/dask/array/backends.py
new file mode 100644
index 00000000000..55de0d17987
--- /dev/null
+++ b/dask/array/backends.py
@@ -0,0 +1,33 @@
+from .core import tensordot_lookup, concatenate_lookup
+
+
+@tensordot_lookup.register_lazy('cupy')
+@concatenate_lookup.register_lazy('cupy')
+def register_cupy():
+    import cupy
+    concatenate_lookup.register(cupy.ndarray, cupy.concatenate)
+    tensordot_lookup.register(cupy.ndarray, cupy.tensordot)
+
+
+@tensordot_lookup.register_lazy('sparse')
+@concatenate_lookup.register_lazy('sparse')
+def register_sparse():
+    import sparse
+    concatenate_lookup.register(sparse.COO, sparse.concatenate)
+    tensordot_lookup.register(sparse.COO, sparse.tensordot)
+
+
+@concatenate_lookup.register_lazy('scipy')
+def register_scipy_sparse():
+    import scipy.sparse
+
+    def _concatenate(L, axis=0):
+        if axis == 0:
+            return scipy.sparse.vstack(L)
+        elif axis == 1:
+            return scipy.sparse.hstack(L)
+        else:
+            msg = ("Can only concatenate scipy sparse matrices for axis in "
+                   "{0, 1}.  Got %s" % axis)
+            raise ValueError(msg)
+    concatenate_lookup.register(scipy.sparse.spmatrix, _concatenate)
diff --git a/dask/array/core.py b/dask/array/core.py
index bdb31a8334c..accf963f636 100644
--- a/dask/array/core.py
+++ b/dask/array/core.py
@@ -61,38 +61,6 @@
 tensordot_lookup.register((object, np.ndarray), np.tensordot)
 
 
-@tensordot_lookup.register_lazy('cupy')
-@concatenate_lookup.register_lazy('cupy')
-def register_cupy():
-    import cupy
-    concatenate_lookup.register(cupy.ndarray, cupy.concatenate)
-    tensordot_lookup.register(cupy.ndarray, cupy.tensordot)
-
-
-@tensordot_lookup.register_lazy('sparse')
-@concatenate_lookup.register_lazy('sparse')
-def register_sparse():
-    import sparse
-    concatenate_lookup.register(sparse.COO, sparse.concatenate)
-    tensordot_lookup.register(sparse.COO, sparse.tensordot)
-
-
-@concatenate_lookup.register_lazy('scipy')
-def register_scipy_sparse():
-    import scipy.sparse
-
-    def _concatenate(L, axis=0):
-        if axis == 0:
-            return scipy.sparse.vstack(L)
-        elif axis == 1:
-            return scipy.sparse.hstack(L)
-        else:
-            msg = ("Can only concatenate scipy sparse matrices for axis in "
-                   "{0, 1}.  Got %s" % axis)
-            raise ValueError(msg)
-    concatenate_lookup.register(scipy.sparse.spmatrix, _concatenate)
-
-
 class PerformanceWarning(Warning):
     """ A warning given when bad chunking may cause poor performance """
 
diff --git a/dask/array/utils.py b/dask/array/utils.py
index 7f289ff4d26..408ef820581 100644
--- a/dask/array/utils.py
+++ b/dask/array/utils.py
@@ -22,7 +22,7 @@
 
 
 def normalize_to_array(x):
-    if 'cupy' in str(type(x)):
+    if 'cupy' in str(type(x)):  # TODO: avoid explicit reference to cupy
         return x.get()
     else:
         return x
diff --git a/dask/dataframe/__init__.py b/dask/dataframe/__init__.py
index 396875f7a4d..ee6c0bbbf95 100644
--- a/dask/dataframe/__init__.py
+++ b/dask/dataframe/__init__.py
@@ -11,7 +11,7 @@
                      read_fwf)
     from .optimize import optimize
     from .multi import merge, concat
-    from . import rolling
+    from . import rolling, backends
     from ..base import compute
     from .reshape import get_dummies, pivot_table, melt
     from .utils import assert_eq
diff --git a/dask/dataframe/backends.py b/dask/dataframe/backends.py
new file mode 100644
index 00000000000..b400395bab6
--- /dev/null
+++ b/dask/dataframe/backends.py
@@ -0,0 +1,37 @@
+from .methods import concat_dispatch
+from .core import get_parallel_type, meta_nonempty, make_meta
+
+
+######################################
+# cuDF: Pandas Dataframes on the GPU #
+######################################
+
+
+@concat_dispatch.register_lazy('cudf')
+@get_parallel_type.register_lazy('cudf')
+@meta_nonempty.register_lazy('cudf')
+@make_meta.register_lazy('cudf')
+def _register_cudf():
+    import cudf
+    import dask_cudf
+    get_parallel_type.register(cudf.DataFrame, lambda _: dask_cudf.DataFrame)
+    get_parallel_type.register(cudf.Series, lambda _: dask_cudf.Series)
+    get_parallel_type.register(cudf.Index, lambda _: dask_cudf.Index)
+
+    @meta_nonempty.register((cudf.DataFrame, cudf.Series, cudf.Index))
+    def _(x):
+        y = meta_nonempty(x.to_pandas())  # TODO: add iloc[:5]
+        return cudf.from_pandas(y)
+
+    @make_meta.register((cudf.Series, cudf.DataFrame))
+    def _(x):
+        return x.head(0)
+
+    @make_meta.register(cudf.Index)
+    def _(x):
+        return x[:0]
+
+    concat_dispatch.register(
+        (cudf.DataFrame, cudf.Series, cudf.Index),
+        cudf.concat
+    )
diff --git a/dask/dataframe/core.py b/dask/dataframe/core.py
index ac4088d1ae2..4c38ce8cd36 100644
--- a/dask/dataframe/core.py
+++ b/dask/dataframe/core.py
@@ -4495,30 +4495,6 @@ def get_parallel_type_frame(o):
     return get_parallel_type(o._meta)
 
 
-@get_parallel_type.register_lazy('cudf')
-@meta_nonempty.register_lazy('cudf')
-@make_meta.register_lazy('cudf')
-def _register_cudf():
-    import cudf
-    import dask_cudf
-    get_parallel_type.register(cudf.DataFrame, lambda _: dask_cudf.DataFrame)
-    get_parallel_type.register(cudf.Series, lambda _: dask_cudf.Series)
-    get_parallel_type.register(cudf.Index, lambda _: dask_cudf.Index)
-
-    @meta_nonempty.register((cudf.DataFrame, cudf.Series, cudf.Index))
-    def _(x):
-        y = meta_nonempty(x.to_pandas())  # TODO: add iloc[:5]
-        return cudf.from_pandas(y)
-
-    @make_meta.register((cudf.Series, cudf.DataFrame))
-    def _(x):
-        return x.head(0)
-
-    @make_meta.register(cudf.Index)
-    def _(x):
-        return x[:0]
-
-
 def parallel_types():
     return tuple(k for k, v in get_parallel_type._lookup.items()
                  if v is not get_parallel_type_object)
diff --git a/dask/dataframe/methods.py b/dask/dataframe/methods.py
index c8f2c2ed0e3..49f7f9b5b56 100644
--- a/dask/dataframe/methods.py
+++ b/dask/dataframe/methods.py
@@ -247,15 +247,6 @@ def _get_level_values(x, n):
 concat_dispatch = Dispatch('concat')
 
 
-@concat_dispatch.register_lazy('cudf')
-def register_cudf():
-    import cudf
-    concat_dispatch.register(
-        (cudf.DataFrame, cudf.Series, cudf.Index),
-        cudf.concat
-    )
-
-
 def concat(dfs, axis=0, join='outer', uniform=False, filter_warning=True):
     """Concatenate, handling some edge cases: