This paper presents a novel architecture for end-
to-end design automation, facilitating high-level design portability across diverse technologies. We introduce programmatic, customizable and reusable design-flows capable of generating multiple implementations (e.g., CPU, GPU, FPGA) from a single technology-agnostic high-level application source. Notably, our approach incorporates design-flow branch points and automated
path selection strategies, mitigating the manual effort currently needed for efficient design production, particularly for heterogeneous platforms. To validate our approach, we implement optimizing design-flows tailored to different hardware platforms. Through experiments on five AI and HPC benchmarks, we demonstrate significant speed improvements compared to single-threaded CPU execution. Our approach generates multi-thread CPU, CPU+FPGA, and CPU+GPU designs from a single high-level source description, achieving speedups of up to 30 times for OpenMP multi-thread CPU, 32 times for oneAPI CPU+FPGA,
and 779 times for HIP CPU+GPU designs. We also showcase cost-effective implementations targeting heterogeneous computing platforms. Additionally, these performance advancements are accompanied by gains in developer productivity, quantified based on added lines of code.