Although Markov state models have proven to be powerful tools in resolving the complex features of biomolecular kinetics, the discretization of the conformational space has been a bottleneck since the advent of the method. A recently introduced variational approach, which uses basis functions instead of crisp conformational states, opened up a route to construct kinetic models in which the discretization error can be controlled systematically. Here, we develop and test a basis set for peptides to be used in the variational approach. The basis set is constructed by combining local residue-centered kinetic modes which are obtained from kinetic models of terminally blocked amino acids. Using this basis set, we model the conformational kinetics of two hexapeptides with sequences VGLAPG and VGVAPG. Six basis functions are sufficient to represent the slow kinetic modes of these peptides. The basis set also allows for a direct interpretation of the slow kinetic modes without an additional clustering in the space of the dominant eigenvectors. Morever, changes in the conformational kinetics due to the exchange of leucine in VGLAPG to valine in VGVAPG can be directly quantified by comparing histograms of the basis set expansion coefficients.