I am looking at https://github.com/iovisor/ubpf as a potential candidate for an eBPF runtime to add to the existing cBPF one.
It is Apache2.0 licensed so IMHO it should be OK to use it "as is" to generate JIT code and/or create an interpreter based on it.
The idea is to implement some of the basic BPF processing + some of BESS specific helpers (get/set metadata) and map well defined generic helpers such as f.e. the map lookups onto BESS ones (f.e. internal hashmap support).