Next-generation cyber physical systems (CPS) face a variety of challenges in terms of networking, interoperability and scalability in the presence of heterogeneous devices. Recent advances in software-defined networking (SDN) make it a viable approach towards addressing these issues by introducing programmability and flexibility in the network. However, the elasticity of scale required by CPS raises some concerns about the control-overhead of the OpenFlow protocol. In this paper, we propose a dynamic traffic engineering scheme, DynamiTE, in software-defined cyber physical systems (SD-CPS). Our aim is to minimize the control overhead at the SDN controller by minimizing the number of PACKET-IN messages. We propose a greedy heuristic approach to determine the optimal number of switches required to have higher ternary content-addressable memory (TCAM), termed as candidate switches, compared to the other switches in the network. In such a scenario, a fully occupied switch directly forwards a new incoming flow to a candidate switch without sending a PACKET-IN message to the SDN controller. Further, a packet-tagging method is applied to notify the SDN controller about the congestion occurred at the fully occupied switch. Simulation results show that DynamiTE is capable of reducing the number of PACKET-IN messages by 10% compared to the OpenFlow-based reactive forwarding schemes (OFS). Further, the number of packets experiencing congestion in the network is reduced by 38%, compared to the randomized forwarding scheme (RFS).