Modern flexible manufacturing systems can usually complete a variety of operations that are needed by multiple products, and the cycle time of the machines may be uncertain. However, traditional discrete-time Markov Chain-based models usually assume the cycle time of machines are fixed. In this paper, we develop a model to analyze the system dynamics and evaluate performance of a two-stage manufacturing system that has multiple parallel machines in each stage and multiple operations assigned to every machine. It is assumed that the time to complete each operation follows a geometric distribution, so the cycle time of each machine follows a discreate phase-type distribution. Simplified one-parameter and two-parameter models are also developed to approximately analyze the performance, and their relations to the traditional models are discussed. Numerical examples are provided to illustrate the proposed models and verify their effectiveness.