This paper studies interconnected networked control systems in the presence of communication constraints and plant uncertainties. We propose a novel control architecture, where a reference model is introduced in each subsystem with the objective of generating the data for network broadcast as well as local feedback control. With this structure, the co-design of an output-feedback control strategy and the required communication protocol can be decoupled: the local controller only needs to address the uncertainty in the system dynamics, while the communication protocol purely focuses on the limitations due to communication channels. As a result, one can integrate the existing communication protocols and robust control techniques into this architecture without worrying about the coupling between the control and the communication. We provide stability conditions for the overall closed-loop system and derive uniform bounds on the error signals between the real system, the reference model, and an ideal model, which has perfect communication and no uncertainties. It is shown that these bounds can be arbitrarily improved, subject only to the hardware limitations.