Cyber-Physical Systems (CPSs) are the backbone of essential services such as critical infrastructural services, smart cities, and industrial automation. It throws concerns around real-time decision making, security breaches, and resource optimization that it is now facing more and more. The paper introduces a new proposition, a Recurrent Integrated Convolutional Neural Network Gate (RICG) model: a dynamic architecture, deep-learning based, useful in increasing security and efficiency in a CPS environment. The RICG model brings together the spatio-feature extraction facility of Convolutional Neural Network (CNN) with the temporal learning power of Recurrent Neural Networks (RNN) to function as an adaptive decision-making mechanism. An integrated gating structure dynamically regulates the way extracted spatial-temporal features flow within the model, hence prioritizing the security-critical data and filtering potential anomalies. Unlike other traditional models, RICG employs context-aware recurrent gates, which adjust information persistence intelligently, making it significantly effective for detecting cyber threats such as intrusion attempts, unauthorized access, and data tampering in real-time. The model also boosts the performance exhibited for the system as it predicts workload patterns, thus balancing resource allocation and significantly reducing latency. Extensive simulations on benchmark CPS datasets showed that the RICG model is by far the best among all typically deep learning methods for providing much better accuracy in less time and equally tests more in terms of robustness in detecting anomalies under even quite dynamic network conditions. This research, thus, opens a vast and smart horizon for the future of intelligent industrial systems and critical infrastructure protection as it nurtures a server framework of secure CPS while it continues to keep the very high efficiency promises. Thus, it heralds the proposed RICG model as a next step towards CPS, autonomously, self-protected on-the-fly reliable decision making even in the most complex ever-evolving threat landscapes.