The growing adoption of cloud computing and Software-as-a-Service (SaaS) platforms has intensified the demand for scalable and performance-isolated multi-tenant database (MTDB) architectures. Traditional partitioning schemes, including hash, range, and hybrid methods, remain static and fail to address the challenges of heterogeneous workloads, workload skew, and elastic scalability. This paper proposes DynaPartMT, a dynamic, workload-aware data partitioning model for MTDBs in cloud platforms. The framework integrates continuous workload monitoring, lightweight forecasting, and a multi-objective cost-aware partitioning planner that jointly minimizes cross-partition transactions, balances load, enforces tenant-level isolation, and bounds migration overhead. The model is mathematically formalized, with optimization objectives and constraints explicitly defined to guarantee scalability and SLA compliance.