Designing a heuristic algorithm to solve an optimization problem can also be seen as an optimization problem. Such a problem seeks to determine the best algorithm contained in the search space. The objective function corresponds to the computational performance of the algorithm measured in terms of computational time, complexity, number of instructions or number of elementary operations. The automatic design of algorithms has been explored for several combinatorial optimization problems. In this work, we extend this exploration towards the automatic design of metaheuristics to find solutions for the traveling salesman problem. The process is carried out by genetic programming. The resulting algorithms are combinations of well-known metaheuristics and, in some cases, present better computational performance than the existing algorithms for the set of selected test instances.