Reinforcement Learning using Kernel-Based Stochastic Factorization