Introduction of new, more advanced services to the networking paradigm has led to an increased heterogeneity of media types and network traffic. Although several transport protocols have been developed over the years to cater to the Quality-of-Service requirements of these network services, the dynamic nature of the network condition is a variable that creates a hindrance in the mapping of an efficient transport protocol to a network service.

Dynamic Protocol Selection can serve as a potential solution for this by applying innovative techniques to adaptively select among pre-existing transport protocols during run-time, thus catering to these requirements dynamically. Although the concept has been proven to be beneficial, there are certain research gaps that are yet to be addressed. In this thesis, we attempt to take a step forward to address a few of these research gaps. As a result, we designed a standardized conceptual framework (DPS framework) for the Dynamic Selection of various TCP congestion control algorithms (as the pre-existing transport protocols). To enable this framework to function autonomously, we developed an online learning strategy implemented in the framework design. Also, to ensure efficient deployability of the framework in a real-network, we further proposed a fairness framework to manage multiple DPS framework-enabled application flows to co-exist sustainably.

Through our experiments, we demonstrated the trade-off between application flow performance and learning time for the proposed online learning strategy and the ways it can be tuned to benefit certain types of application flows. We further presented results that showcased around $15\%$ improvement in the fairness performance between multiple application flows and stability in the average flow performance due to the implementation of the proposed fairness framework.