Device-to-Device (D2D) communications enable cellular devices in physical proximity to communicate directly over the air instead of through the BSs, which can be applied in Intelligent Transportation Systems (ITS) to efficiently support the communications between vehicles. However, the resource control functions need to adapt to the unique characteristics of vehicular environments and meet the high reliability and low latency requirements of road safety applications. In this paper, we discuss the performance evaluation and dynamic optimization of D2D communications for ITS based on stochastic models, where the complex mobility behavior of vehicles and stochastic nature of traffic arrivals are captured. We focus on two typical scenarios, i.e., safety applications and vehicular cloudlet, and outline the research directions and present the challenges and major issues that need to be addressed.