Why should I consider models like YOLO, SSD, Faster R-CNN ,when TensorFlow or OpenCV are available options for custom object detection without models

Question

Is it possible to create a custom object detection system without utilising established models like YOLO, SSD, Faster R-CNN, and similar approaches? If so, what are the steps to implement it? If not, what are the reasons behind its in-feasibility?

I've experimented with implementing custom object detection using YOLO in conjunction with OpenCV and SSD with Tensorflow, and I find myself questioning the reasons for using these models.