Bounds and Low-rank Approximation for Controlled Markov Processes