Cognitive Radio Networks (CRNs) offers an effective solution for radio spectrum scarcity problem and cyber-security to the growth of 5G technologies and the Internet of Things (IoT) devices. In this paper, the CRNs do cross-layer (X-layer) optimization by finding the parameters from the physical layer (Layer-1) of the OSI model and the network layer (Layer-3) of the OSI model so as to progress the end-to-end secure transmission of cognition for 5G and IoT traffic. The proposed model follows the invention target by applying a Deep Q-Network (DQN) to select after that hop for sending based on the waiting duration found on every router when keeping SINR lower than threshold determines by primary channel. A fully linked feed-forward Multilayer Perceptron (MLP) model is applied by secondary users (SUs) to estimate the activity value function. The activity value contains SINR to the primary user (PU) at the layer-1 and following hop to the routers for every packet at the layer-3. The advantage to the neural network (NN) is Mean Opinion Score (MOS) for secure encrypted high-resolution video traffic over 5G network which depends upon the packet drop rate and the bit error rate applied for transmission. As evaluated to the execution of DQN learning at the physical layer, this system gives for 37% gain in the video quality for routers with short queues and besides reaches a balanced load upon a network with routers with different service rates.