Using the lifting technique to discretize the non-uniformly sampled nonlinear system into a multiple input single output transfer function model, the system output can be represented as a linear parameter model with non-uniform updating nonlinear input and ouput regressors, and the system can be further identified based on the estimates of nonlinear inputs or the overparameterization method. However, when the nonlinear structure is unknown or cannot be parameterized by the measurable non-uniform inputs, the above mentioned identification methods will no longer apply. In order to solve this problem, the kernel method is applied to project original nonlinear data into a high dimensional feature space to make it linearly separable, and then the recursive least squares algorithm is used to identify the projected data, thus a kernel recursive least squares identification algorithm is proposed for non-uniformly sampled nonlinear systems. In addition, refering to the idea of recursive extened least squares algorithm, replacing the unmeasured noise with the estimated residual error, the kernel recursive extended least squares algorithm is proposed for systems with colored noise interference. Finally, numerical simulation examples verify the effectiveness of the proposed algorithms.