Yes.
Do you want y1 and y2 to be equally weighted? If so, then composite can be computed with the usual equations for sum of normal distributed random variables (Sum of normally distributed random variables - Wikipedia)
kfold_elpd_composite <- kfold1$estimates["elpd_kfold","Estimate"] + kfold2$estimates["elpd_kfold","Estimate"]
kfold_elpd_se_composite <- sqrt(kfold1$estimates["elpd_kfold","SE"]^2+kfold2$estimates["elpd_kfold","SE"]^2)
and corresponding equations can be used also for the results from loo_compare (which works also for kfold objects)