Disentangling The Octopus: Towards Decomposing Reinforcement Learning Systems In To Intuitive Subsystems