Performance tuning is a pervasive challenge in science and engineering, involving optimization of expensive black-box functions to achieve desired outcomes. This thesis focuses on Bayesian optimization (BO) as a promising solution for addressing the complexities of performance tuning. It proposes a series of principled Bayesian optimization algorithms in scenarios with black-box constraints, prior structural knowledge, and human-in-the-loop feedback. Key contributions include: (1) a constrained BO algorithm with theoretical performance guarantees, applied to tune the performance of a building controller and reduce building electricity costs by up to 27% while maintaining comfort constraints; (2) a scalable distributed BO algorithm leveraging additive structures; (3) human-in-the-loop BO algorithms that optimizes human preferences; and (4) human-AI collaborative BO that exploits expert supervisory feedback to enhance the convergence speed. These advancements address practical challenges in performance tuning, expanding the applicability of Bayesian optimization to real-world problems.