$\newcommand\de\delta$The maps $$\mu\mapsto\sqrt{\mathrm{KL}(\mu\|\nu)}$$ and $$\nu\mapsto\sqrt{\mathrm{KL}(\mu\|\nu)}$$ are not convex in general. Indeed, let $\mu_p:=p\de_0+(1-p)\de_1$, where $p\in(0,1)$ and $\de_a$ is the Dirac measure supported on $\{a\}$. Then the second partial derivative with respect to $p$ of $\sqrt{\mathrm{KL}(\mu_p,\mu_r)}$ at $(p,r)=(1/10,1/11)$ is $-7.17\ldots<0$. So, $\sqrt{\mathrm{KL}(\mu,\mu_r)}$ is not convex in $\mu$. Also, the second partial derivative with respect to $r$ of $\sqrt{\mathrm{KL}(\mu_p,\mu_r)}$ at $(p,r)=(1/10,1/9)$ is $-11.50\ldots<0$. So, $\sqrt{\mathrm{KL}(\mu_p,\nu)}$ is not convex in $\nu$. --- You also asked: "If this is not true in general, does it exists a measure $\mu\in P(X)$ such that $\nu\mapsto \sqrt{\mathrm{KL}(\mu\|\nu)}$ is a convex map $P(X)\to \mathbb [0,+\infty]$? Or a measure $\nu$ such that $\mu\mapsto \sqrt{\mathrm{KL}(\mu\|\nu)}$ is convex?" The answer to each of these two questions is yes, at least when $X=\{0,1\}$, say. For $p\in(0,1)$, let \begin{equation} F(p):=\sqrt{\mathrm{KL}(\mu_p,\mu_{1/2})}, \end{equation} \begin{equation} f(p):=F''(p)4 \mathrm{KL}(\mu_p,\mu_{1/2})^{3/2}, \end{equation} \begin{equation} f_1(p):=f'(p)(1-p)^2 p^2. \end{equation} Then $f_1(1/2)=f'_1(1/2)=f''_1(1/2)=0$ and \begin{equation} f'''_1(p)=\frac{2+4 p(1-p)}{(1-p)^2 p^2}>0. \end{equation} It follows that $F''(p)\ge0$, so that $\sqrt{\mathrm{KL}(\mu,\mu_{1/2})}$ is convex in $\mu$. For $r\in(0,1)$, let \begin{equation} G(r):=\sqrt{\mathrm{KL}(\mu_{1/2},\mu_r)}, \end{equation} and \begin{equation} g(r):=G''(r)4\mathrm{KL}(\mu_{1/2},\mu_r)^{3/2}. \end{equation} Then $g(1/2)=g'(1/2)=g''(1/2)=g'''(1/2)=0$ and \begin{equation} g''''(1/2+h)\frac{(1 - 4 h^2)^4}{16}=9- 16 h^4 + 156 h^2 + 64 h^6 >9- 1 + 156 h^2 + 64 h^6>0 \end{equation} if $|h|<1/2$. It follows that $G''(r)\ge0$, so that $\sqrt{\mathrm{KL}(\mu_{1/2},\nu)}$ is convex in $\nu$.