It is known that analytic modelling for TCP latency is a non trivial task. Recently, some significant progress has been made, such as the comprehensive result by Sikdar et al. However, models similar to these often rely on trial and error methods such as "data fitting". This can lead to a very limited scope for the resulting model and also large estimation error. In this paper, we propose improvements to Sikdar?s SACK TCP model. A new delayed acknowledgement slow start model is developed that is analytically derived from the slow start algorithm which provides a novel mechanism to model the relationship between RTT and the delayed acknowledgement timer. We introduce a simple mechanism to include time taken to send an HTTP Get Request to broaden the scope of our SACK TCP model to website file transfer. Simulation and live Internet experimentation has validated our scheme.