Model factories are stand alone functions which take an arbitrary number of primitive parameters (int, float, list, dict, etc) and return a model which can then be used in the kind parameter of some Scikit-Learn like wrapper model.
An example of this is KerasAutoEncoder which accepts a kind argument (as all custom gordo models do) and can be given feedforward_model. Meaning that function will be used to create the underlying Keras model for KerasAutoEncoder
gordo.machine.model.factories.feedforward_autoencoder
gordo.machine.model.factories.lstm_autoencoder