In one of the several manifestations, the future cellular networks are required to accommodate a massive number of devices; several orders of magnitude compared to today's networks. At the same time, the future cellular networks will have to fulfill stringent latency constraints. To that end, one problem that is posed as a potential showstopper is extreme congestion for requesting uplink scheduling over the physical random access channel (PRACH). Indeed, such congestion drags along scheduling delay problems. In this paper, the use of self-organized device-to-device (D2D) clustering is advocated for mitigating PRACH congestion. To this end, the paper proposes two D2D clustering schemes, namely; Random-Based Clustering (RBC) and Channel-Gain-Based Clustering (CGBC). Accordingly, this paper sheds light on random access within the proposed D2D clustering schemes and presents a case study based on a stochastic geometry framework. For the sake of objective evaluation, the D2D clustering is benchmarked by the conventional scheduling request procedure. Accordingly, the paper offers insights into useful scenarios that minimize the scheduling delay for each clustering scheme. Finally, the paper discusses the implementation algorithm and some potential implementation issues and remedies.