A Performance Focused, Development Friendly And Model Aided Parallelization Strategy For Scientific Applications