Emergence of Artificial Intelligence (AI) assisted Internet of Things (IoT) enables cognitive functions with integration of massive sensors and actuators nodes over a ubiquitous cyber-infrastructure making Next-gen Cyber Physical System (NG-CPS) a reality. The payload is the ‘sensor data’ encapsulated and carried by the TCP/IP protocol data units (PDUs) in IoT and CPS applications. Information Gain (IG) of all the data-points are not same in time-series sensor data. IG can help in understanding the criticality of data and thereby help in improving the Quality of Service (QoS) and network resource allocation for faster transmission of critical data-points. This paper proposes a novel Content-Aware Prioritization Scheme (CAPS) for multi-class traffic profiling, queuing and scheduling based on multi-variate non-linear features from sensor data for fixing priority to enhance the QoS of critical sensor data. As proof of concept, real time prediction of priority of sensor data is done by Machine Learning (ML) model that has been trained and tested over historical sensor data. We have used various classification algorithms such as SVM, Naive Bayes, and ANN-based Multi-layer Perceptron (MLP) for predicting four priority classes of real-time sensor data based on multi-variate features of the sensor data. A performance comparison results of these models shows 93% highest accuracy for real time prediction of priority of sensor data. For enhanced Network Resource Management (NRM), four Class-based Queuing (CBQ) with Weighted Round-Robin (WRR) scheduler are deployed in NS-2 for validation QoS performance using throughput, latency, PDR, and jitter depending on the predicted priority of the sensor data carrying PDUs.