This dissertation develops theory, models, and operational methods for preserving and optimizing spin polarization in hadron accelerators spanning the RHIC injector chain and the Electron-Ion Collider (EIC) Hadron Storage Ring (HSR). Unlike electrons, hadrons do not self-polarize at high energy; source polarization must survive long acceleration ramps that traverse hundreds of depolarizing resonances and multiple machines. The work integrates high-fidelity spin–orbit modeling, snake lattice design, and data-driven beam tuning into a reproducible framework. Key contributions are:(1) Application of a Maxwellian field representation and symplectic tracking algorithm for AGS combined‑function dipoles and partial Siberian snakes, removing non‑symplectic tracking artifacts and enabling stable million‑turn spin-orbit simulations, yielding realistic polarization transmission over most of the AGS cycle. (2) Conception, and optimization of the Doubly Lee–Courant (DLC) six‑snake scheme for the HSR, achieving near‑perfect polarization preservation for (protons and) helions and establishing a resilient baseline configuration. (3) First experimental deployment of Bayesian optimization for injection tuning in the BtA transfer line (Booster → AGS), improving transmission from ≈ 65% → 90+% with no operator intervention. (4) Experimental validation of resonant slow extraction modeling at NSRL, confirming normal‑form predictions and achieving one to two orders of magnitude dispersion reduction. Together these results unify snake design, orbit control, and machine‑learning operations, providing a practical path to EIC polarization goals and a computational foundation for future polarized hadron facilities and their automated optimization.