Vehicular Cloud Computing is a promising solution to exploit the underutilized vehicular resources and to meet the requirements of VANET applications and services. Although modern vehicles have important capacities of computation and storage, there is an increasing need for resources, in particular, for safety applications which require the cooperation between vehicles. The vehicular cloud offers to users the opportunity to rent resources on-demand or to share them freely to run their applications or to carry out some tasks. Even though this paradigm is feasible, its implementation still faces problems. Many researchers have focused on the architectural design in order to overcome different challenges and consequently meet user requirements to provide him/her with reliable services. In this work, we survey the vehicular cloud paradigm. We focus on its features and architectures. We first present a brief overview of the motivation of vehicular cloud. Then, we explore challenges related to its design. Furthermore, we highlight the features of existing vehicular cloud architectures: we provide a taxonomy of vehicular cloud followed by our classification criteria. Finally, we discuss issues that can be considered as open research directions.