This paper presents a constraint-guided architectural framework for generative system design based on Structural Functional Layers (SFL). Building on a material-agnostic functional architecture and SFL taxonomy, the work formalizes a pre-computational translation pipeline that maps system requirements into architecturally valid solution spaces prior to simulation, optimization, or AI-assisted processing. The framework introduces a rule-based exclusion logic (R1–R21) and a relevance criterion for conflict identification (RK-K), ensuring functional separation, interface protection, and architectural consistency. Rather than proposing an AI model or optimization method, the paper defines an upstream constraint infrastructure that reduces search spaces, prevents invalid configurations, and improves explainability for downstream computational tools. The approach is model-agnostic, implementation-independent, and applicable across domains involving complex multi-functional systems.