Estimation of Worst Case Execution Times (WCETs) of software functions in a real-time embedded system is fundamentally important in verifying its timing behaviour. Many existing WCET analysis tools and benchmarks, based on static WCET analysis methods, are limited to analyse software functions that conform to specific programming languages and libraries. As a consequence, these tools do not support WCET estimation of software functions in legacy industrial systems that do not conform to those languages and libraries. This paper advocates the use of a statistical method based on extreme value theory for estimating WCETs of software functions in such legacy industrial embedded systems. The main advantage of this method is that it is agnostic of the languages and libraries that are used to implement the software functions. In order to provide a proof of concept, the paper incorporates an industrial use case and applies the method to estimate WCETs of its software functions. The paper also presents an extensive comparative evaluation of the statistical method and an established static analysis tool for estimating WCETs that utilises high-level flow analysis.