Exploring Optimisations for the Local Assembly phase of Finite Element Methods on GPUs