Abstract : Vehicular Ad Hoc Networks (VANET) provide vehicle-to-vehicle communications to increase the road safety. The safety-related applications have gained a lot of attention. A promising example is the Local Danger Warning application that increase the line-of-sight of the driver by providing warnings about potential dangerous situations ahead. Because of their real-time constraints and life-threatening consequences of misbehavior, communications have to be secured. But, security mechanisms come with processing and network overheads. The main goal of this PhD thesis is to propose a general (and analytical) framework on the impact of the security overhead on the transfer delay of a warning. Among the conventional security mechanisms, the authentication mechanism appears as the cornerstone of VANET security. Moreover, the authentication is used for every message sent or received. So it may be the most consuming mechanism. This is the reason why we first focus on the authentication overhead and propose a global formula to assess it. But the authentication won't be the only security mechanism to be deployed in VANET. The consensus is often used to provide trust between vehicles. Indeed, with a decision method and from a set of messages, the consensus aims at a common agreement on a value or an action between vehicles. We analyze how to define the consensus parameters in order to respect the real-time constraints. The second contribution of this thesis is a dynamic decision method that uses the network environment and the content of the warning to decrease the number of messages to verify and so provide an adapted reaction.