Incorporating vehicle-to-everything (V2X) and device-to-device (D2D) communication in a vehicular environment has become a research focus area due to the proliferation of fifth-generation (5G) technology. However, establishing V2X and D2D communication remains a challenging issue in urban scenarios because, by the nature of their role and usage, vehicles are subject to dynamic changes in their mobility and direction, as well as being affected by traffic congestion. Moreover, safety message dissemination over vehicular ad hoc networks (VANETs) without the introduction of a broadcast storm poses a considerable problem. This paper presents a design for a novel cellular-5G VANET architecture intended to address all prior problems in VANETs and improve quality of service (QoS). To this end, all vehicles in the field in focus are divided into clusters using an adaptive mobility aware path similarity (A-MAPS) algorithm. Cluster formation and cluster head (CH) selection are performed on the basis of a number of pivotal criteria, including future path similarity, which is vital for urban environments. To establish reliable vehicle-to-vehicle (V2V) and D2D communication, a Bayesian rule-based fuzzy logic (BRFL) algorithm is introduced that determines the optimal forwarder for V2V and the optimal device for D2D communication. Furthermore, vehicle-to-infrastructure (V2I) and vehicle-to-pedestrian (V2P) communication are handled with a novel two-fitness hypotrochoid spiral optimization (2F-HSO) algorithm that is derived using fitness functions. For safety message dissemination, two different message types (Accident and Traffic) are considered, and a safety aware hierarchical tree for dissemination (SA-HTD) is constructed to handle their dissemination. The proposed cellular-5G VANET is modeled in an OMNeT++ simulator. Results indicate improvement in packet delivery ratio (PDR), throughput, transmission delay, and dissemination delay.