The emerging scenarios and use cases for the Internet of Things (IoT) and the latest developments of 5G and beyond networks envision a connected world with diverse stakeholders as an integral and dynamic part of it. The synergistic coming together of the different stakeholders also brings along with it different values, requirements, and policies that need to be orchestrated in an automated and agile manner. This presents a requirement for a dynamic and context-aware system that can provide the assimilation of the different policies and requirements of the stakeholders and perform a dynamic orchestration in the system. Artificial intelligence has a major role in bringing in the desired adaptability and automation in the context-aware system. Besides, with the advent of fog computing for IoT and multi-access edge computing (MEC) in 5G, the execution location also introduces benefits and challenges bringing along different engineering trade-offs such as for execution performance, maintainability, and usability.

This thesis aims at exploring and analyzing the engineering trade-offs of employing (artificial intelligence) AI-based and a static rule-based system for these orchestration requirements. It further explores the engineering trade-offs of employing the proposed systems at a cloud-only or fog setup. Scenarios are taken from different industry experts and evolved further. Different architecture solutions are proposed and a scenario-based architecture assessment is performed. A PERT (Program Evaluation Review Technique) analysis is also performed for the systems and the change scenarios for them. Two prototypes are developed using C++ and an expert system (for AI) and measurements are captured and evaluated for the different scenarios and configurations. The performance trade-offs of the execution of the scenarios on the two prototypes while executing over fog or cloud-only setups are evaluated. Execution performance is measured in terms of the time taken for the execution and a comparative analysis is done with graphs and charts explaining the various execution trade-offs. Maintainability and usability trade-offs are also discussed in the light of the two reference systems executing on the two locations, viz. fog, and cloud. As the next-generation IoT and telecommunication systems have diverse and strict quality of service (QoS) requirements, the challenges of learning in the system are evaluated and discussed in the thesis. Bayesian networks and probabilistic programming are explored and evaluated for validating the input of small data for evolving an existing expert system model.