Today s mobile communications systems are facing new challenges, triggered by the increased use of new devices and the growth of bandwidth hungry applications. This is why over the last years, the telecommunication industry and academic communities have been focused on research and development of technologies for the upcoming 5th generation mobile systems (5G). Among the potential candidates, network densification has attracted growing attention as a key mechanism to fulfill the objective proposed in 5G, by increasing the number of radio-base stations (on the coverage area) and introducing an additional layer of low-power access nodes (e.g., Femto, picocells, relay nodes). However, this approach has also posed new challenges in network configuration, management, and optimization tasks to ensure and increase the mobile network efficiency. This research proposes the inclusion of cognitive mechanisms and adaptive techniques in the architectures of mobile radio access networks. This work also proposes new self-organizing (SON) functions for load balancing, enhanced with capabilities of learning from previous experiences to achieve future desired performance goals.