prepare CNN class for filter benchmarking

• be able to set layer configuration from outside as in FCN and RNN classes
• be able to process filtered data (only process as branched input and not with additional dimension to ensure comparibility with other architectures)
• also add new class branched input CNN