Performance models for Network-on-Chip (NoC) are essential for design, optimization and Quality of Service (QoS) assurance. Classical queueing theory has been often used to provide fast analytical models to estimate average performance. This paper presents a new analytical model that focuses on QoS assurance. It assumes that the NoC has an underlying synchronous behavior with constant service time routers. The comparisons with simulation results show a tangible improvement with regard to the classical M/D/1 models when estimating the worst-case latencies and queue delays. The model can be applied to any network modeled as a queueing system with constant-time routers.