Program debloating aims to enhance the performance and reduce the attack surface of bloat applications. Several techniques have been recently proposed to perform program debloating and remove (or secure) dead code from applications. These approaches are either based on unsound strategies or too strict assumptions on the program. Therefore, these approaches are hardly employable in practice. In this paper, we address these limitations by leveraging symbolic execution and partial evaluation to generate specialized applications. Our approach relies on a simple observation that applications comprise configuration and main logic. The configuration specifies which functionalities in the main logic that should be executed. The symbolic execution provides the partial evaluation of a set of variables that will become constants. The partial evaluation then propagates the constant variables, simplifies the application, and finally generates the specialized application. Our evaluation over different sets of real-world applications demonstrates that LMCAS can successfully remove all unwanted features. LMCAS achieves a 25% reduction in the binary size and can reduce the number of functions by 40% in the specialized applications. LMCAS enhances the security of specialized applications. It reduces the attack surface of code-reuse attacks by removing 51.7% of the total gadgets and eliminates 80% of known CSV vulnerabilities. LMCAS can run up to 1500x and 1.2x faster state-of-the-art debloating techniques CHISEL and OCCAM. Further, we demonstrate the scalability and robustness of LMCAS on three large applications: tcpdump, readelf, and objdump. LMCAS is thus a practical approach for debloating real-world programs.