In Cyber-Physical Systems, the software components are often distributed over several computing nodes, connected by a communication network. Depending on several factors, the behavior of these components may dynamically change during its execution. The existing data flow formalisms for the performance prediction of dynamic systems do not cover the real-time constraints of these systems, and suffer from complexity issues in the verification of mandatory model properties. To overcome these limitations, we propose a dynamic extension to Polygraph, a static data flow formalism covering the real-time behavior of the CPS components. We also propose a verification algorithm to determine if the transitions between different modes are well-defined for a given model. Initial experiments show that this algorithm can be efficiently applied in practice.