State Space Decomposition and Subgoal Creation for Transfer in Deep Reinforcement Learning