With the exponential growth in the number of connected devices, in recent years there has been a paradigm shift toward mobile edge computing. As a promising edge technology, it pushes mobile computing, network control, and storage to the network edges so as to provide better support to computation-intensive Internet of Things (IoT) applications. Although it enables offloading latency-sensitive applications at the resource-limited mobile devices, decentralized architectures and diversified deployment environments bring new security and privacy challenges. This is due to the fact that, with wireless communications, the medium can be accessed by both legitimate users and adversaries. Though cloud computing has helped in substantial transformation of global business, it falls short in provisioning distributed services, namely, security of IoT systems. Thus, the ever-evolving IoT applications require robust cyber-security measures particularly at the network's edge, for widespread adoption of IoT applications. In this vein, the classic machine learning models devised during the last decade, fall short in terms of low accuracy and reduced scalability for real-time attack detection across widely dispersed edge nodes. Thus, the advances in areas of deep learning, federated learning, and transfer learning could mark the evolution of more sophisticated models that can detect cyberattacks in heterogeneous IoT-driven edge networks without human intervention. We provide a SecEdge-Learn Architecture that uses deep learning and transfer learning approaches to provided a secure MEC environment. Moreover, we utilized blockchain to store the knowledge gained from the MEC clusters and thereby realizing the transfer learning approach to utilize the knowledge for handling different attack scenarios. Finally, we discuss the industry relevance of the MEC environment.