Artificial intelligence (AI) has enabled a new paradigm of smart applications, such as extended reality (XR), which comprises mobile augmented reality (MAR), mixed (MR), and virtual reality (VR), which have very stringent latency requirements, especially for applications on mobile devices (e.g., smartphones, wearable devices, and autonomous vehicles). Edge computing-assisted mobile AI systems have emerged as effective ways to support computation-intensive and latency-sensitive applications for mobile devices due to the offloading capability of heavy computational burdens. However, the high mobility of users and instability in wireless networks decrease the overall Quality-of-Service (QoS) of an edge-AI application running on mobile devices with non-linear battery discharge properties. This dissertation presents a comprehensive experimental study of mobile AI applications, considering different DNN models and processing sources, focusing on computational resource utilization, delay, and energy consumption. Additionally, a novel Gaussian process regression-based general predictive energy model is proposed based on DNN structures, computation resources, and processors, which can predict the energy for each complete application cycle irrespective of device configurations. In addition, a novel performance analysis modeling framework of XR applications is proposed, considering heterogeneous wireless networks and using experimental data collected from testbeds designed specifically for this research. A comprehensive performance analysis model is challenging to design due to the dependence of the performance metrics on several difficult-to-model parameters, such as computing resources and hardware utilization of XR and edge devices, which are controlled by their operating systems, and the heterogeneity in devices and wireless access networks. These challenges and ways to overcome them are also presented in detail.Following the performance analysis model, performance enhancement of mobile AI applications is also proposed in this research. This dissertation focuses on the high mobility of users in connected and autonomous vehicles (CAVs). Time-sensitive information update services are necessary for these CAV-AI applications to ensure the safety of people and assets, and satisfactory entertainment applications. However, information from roadside sensors and nearby vehicles can get delayed in transmission due to the high mobility of vehicles. This research proposes a novel periodic predictive AoI-based service aggregation method for CAVs, which can process the information updates according to their update cycles by maintaining a satisfactory latency and data sequencing success rate (DSSR) for CAV-AI applications.Furthermore, an unstable wireless network poses a critical challenge for real-time mobile AI applications. An H.264 video encoding-based edge-MAR system is proposed, with a focus on network conditions, resource utilization, detection accuracy, and energy consumption of various mobile devices. This extensive study provides essential guidelines for network- and energy-aware H.264 video encoding-based Edge-MAR system design to overcome the challenges posed by unstable wireless networks.Finally, a novel deep reinforcement learning-based smart edge-MAR system -- Reinforced Edge-Assisted Learning (REAL), is proposed in this research, where the edge server is exploited to unleash its potential by providing smart and dynamic MAR processing decisions to the devices based on dynamically changing system states, such as wireless link qualities and battery energy levels of mobile devices. The novelty of REAL lies in solving the complex state-transition problem in a stochastic environment through online soft actor-critic learning and delivering reward-based actions to mobile devices to improve the end-to-end latency, energy consumption, accuracy, and offloaded data size collectively.