Some files in symbolic circuits are growing to a large number of lines - for instance layers, parameters, etc. It would make sense to split them into submodules (as also done in the torch module) so that it is easier to matain the code base. It could be nice to have a base.py file containing all the abstractions (e.g. layers/base.py) and other files implementing those abstractions for different purposes (e.g. layers/input/probabilistic.py for input layers that model probabilistic inputs).